Qwen 3.8 27B available on Cerebras at 1500 tokens/s
- ID
- 21259
- Status
- summarized
- Published
- 04 Sep 2026, 2:32 AM
- Fetched
- 04 Sep 2026, 6:14 AM
- Provider
- Hacker News
- Category
- dev-community
- Original URL
- https://inference-docs.cerebras.ai/models/overview
- Source URL
- https://hnrss.org/best
Summary
- Score
- 7.0
- Created
- 04 Sep 2026, 6:15 AM
- Tags
- Audience
- developersai_ml_learnersai_agent_usersvibe_coders
What happened
Cerebras Inference now lists Qwen 3.8 27B (~1500 tokens/s, 64k/128k context) and OpenAI GPT-OSS 120b (~3000 tokens/s, 65k/131k context) on their public endpoints with free trial and pay-as-you-go tiers. All public models are unpruned originals using selective weight-only quantization in storage, with activations and KV cache kept at full precision.
Why it matters
If you're building AI agents or LLM pipelines that are bottlenecked on inference latency, Cerebras offers 3-10x faster token throughput than typical GPU-based API providers for these specific open-weight models. Evaluate whether switching your API endpoint to Cerebras for Qwen 3.8 27B or GPT-OSS 120b materially reduces agent round-trip times before committing.
Discussion angle
Compare real-world cost-per-million-tokens and effective latency for agent workflows between Cerebras and your current provider — the headline tokens/s only matters if rate limits and pricing hold up under your actual load.